►
From YouTube: DevoWorm (2023, Meeting #27): DevoML/DL, Differentiation Genomics, Constructive Connectomics
Description
GSoC updates, Developmental ML/DL and improving DevoLearn. Differentiation Genomics, Cell Fates, and Epigenetic Switches. June 2023 OpenWorm Newsletter. New directions in constructive connectomics. Attendees: Sushmanth Reddy Mereddy, Bradly Alicea, Susan Crawford-Young, Richard Gordon, Jyothi Swaroop, Lukas, Sai Harshitha, and Jesse Parent.
A
B
Are
you
and
dick.
C
Right
now
this
week,
I
haven't
done
any
pretty
work,
planning
actually
my
tone
to
stop
the
work
and
read
one
book
which
he
have
shared
I'm
going
through
that
book
like
building
a
better
project.
It
is
mostly
about
building
a
better
project
using
silent
and
other
softwares
yeah
I'm
reading
that
book
right
now.
Actually,
a
couple
of
days
back,
you
have
shared
me,
some
materials
of
C
elegans.
C
C
C
B
So
I
guess
the
way
to
read.
This
is
not
to
like
just
read
everything
at
once,
like
a
lot
of
the
papers
are
kind
of
examples
of
things
where
people
have
used
machine
learning
for
C,
elegans
or
specifically,
deep
learning.
So
I
wanted
to
give
you
some
papers
on
things
that
people
have
kind
of
done
in
the
past,
but
also
in
development.
So
they
have
a
lot
of
the
things
like
there
will
be
a
lot
of
terms
like
lineage
trees,
and
you
know
different
places
in
the
different
cells
in
the
embryo.
B
B
But
it's
it's
something
that
you'll
need
to
learn
to
know
kind
of
you
know
how
like
at
least
in
C
elegans,
how
a
lot
of
the
terminology
for
development
like
for
the
lineage
tree,
analyzes
that
they
do
and
they
do
what.
Sometimes
they
do.
Gene
expression
analysis,
which
is
even
harder
in.
B
B
C
I
understood
about
it
a
little
bit
right
now.
There
is
a
book
also
and
Ultra.
I
was
pretty
confused
from
there
to
start
paper,
but
yeah.
E
C
Will
read
them
all
for
literature,
review
and
I
will
Implement
My
Method
and
then
I
will
write.
What
is
difference
from
my
mother
and
other
methods
that
could
be
fine,
I
think
so,
and
yeah
I
was
like
pretty
stuck
up
on
test
cases.
Writing
test
case
for
the
moral
Mike
told
that
he
will
hop
into
a
one
meeting
and
he
will
explain
me
how
to
do
that.
C
B
I
I
have
I
mean
you
know,
that's
something
that
I
think
it's
coming
along:
fine
yeah.
We
had
that
meeting
this
last
week
where
we
talked
about
a
lot
of
the
progress
that
you're
making,
and
so
that's
good.
C
Reading
those
papers-
and
he
was
trying
to
help
me
with
the
manuscript
and
the
accuracy
of
the
model,
actually
comparing
different
model
so
yeah,
my
main
motto-
is
To
Ride
with
Jyoti
me
you
mayuk
and
minus
team
of
writing.
This
whole
thing
from
this
thing
I
have
made
and
we
started
work.
Actually
we
divided
some
papers
between
us
between
both
of
us.
You
will
read
some
papers
and
I
will
read
some
papers,
whatever
I
understood
from
my
previous
I'll.
Explain
it
to
him
and
whatever
he
understands,
he
will
explain
to
me
in
this
way.
C
We
can
easily
we
differ
work
and
do
the
work
fastly
area
I
was
thinking.
Go
like
this,
we'll
try
to
complete
as
soon
as
possible.
All
those
scriptures
and
yeah
I
will
start
writing
the
abstract
and
all
this
stuff.
B
Yeah,
so
hello
Josie,
how
are
you
he's
there.
B
B
It's
okay,
so
yeah.
Well,
thanks
for
the
updates
this
month
and
yeah,
let
me
go
over
the
papers
then,
and
we'll
talk
about
some
of
the
other
things
like
you
know
what
we
might
do
so
such
month
and
and
I
met
I,
don't
know!
Last
week
sometime
we
talked
about
the
sort
of
the
project
going
forward
and
then
some
of
the
things
that
he
wants
to
put
out
a
couple
of
Publications
in
a
pipeline
so,
like
you
know,
he's
got
the
idea
of
he
wants
to
work
on.
B
So
we
have
a
paper
which
is
a
kind
of
a
pre-print.
That's
a
very
Broad
View
overview
of
Devo
Divo
learn
and
then
he
wants
to
work
on
improving
that,
which
is
something
we're
trying
to
do
last
year,
but
we
didn't
get
to
it
and
so
that'll
that'll
be
something
that
will
flesh
out
more.
B
Then
he
wants
to
do
a
paper
on
like
benchmarking,
which
is
I,
think
a
good
paper
for
if
you
were
to
submit
to
a
conference
or
a
journal,
you
know
they
always
want
to
have
benchmarks
for
your
your
software
to
make
sure
that
it's
performs.
You
know
compared
to
something
else
and
then
another
paper
which
I
don't
remember
exactly
what
it
was.
B
But
let's
come,
let's
see
him
yeah,
so
this
has
been
presenting
Med
Sam,
which
is
the
Sam
model
segment,
anything
model,
but
like
sort
of
for
medical
applications,
but
then
we'd
have
like
a
version
that
would
do
things
for
developmental
cells.
So
it's
it's
like
it's
like.
You
know,
take
off
on
that
and
then
put
it
writing
that
up.
So
that's
good!
B
B
This
is
for
geophy's
benefit
about
how
to
write
a
paper,
and
you
know
the
way
we
usually
approach
it
is.
You
know
the
classic
introduction
methods,
results
and
discussion
format.
You
have
the
introduction,
where
you
state
the
thing
that
you're
doing
maybe
the
problems
that
you're
going
to
talk
about.
B
Then
you
move
on
to
the
methods
which
are
you
know,
all
the
methods
that
you
use
and
then
the
results
which
may
be
like
your
benchmarking
are,
like
you
know
something
like
a
screenshot
where
you're
putting
in
like
this
is
what
our
software
looks
like
and
then
you
know
discussion
which
doesn't
have
to
be
long.
What
sometimes
can
be
long
which
discusses
how
you
might
use
it?
You
might
also
have
a
use
cases
section
where
you
talk
about
maybe
one
or
two
use
cases
where
you're
applying
the
software
different
problems.
B
So
if
you
were
to
say
we
could
apply
this
to
c
elegans
embryo,
but
maybe
also
to
other
embryos
you
know
it'd
have,
it
would
be.
It
was
a
different
problem,
so
you
have
to
kind
of
spell
it
out
for
people
how
to
use
these
benchmarks
or
the
software
or
whatever
The
Main
Idea.
Here
is
we
want
to
make
you
know
each
paper
would
be
for
a
different
audience
of
readers
so
like
if
you
wanted
to
just
know
something
about
the
software.
B
Maybe
you
weren't,
like
a
computer
scientist,
even
the
the
overview
paper
I
think
would
be
good
because
you
can
just
kind
of
give
people
an
idea
of
what
it
is
and
how
to
use
it,
and
that's
not
going
to
be
the
same
as
like
what
computer
scientists
might
want
to
see,
which
is
a
series
of
benchmarks,
because
they
know
you
know
how
to
use
software.
It's
not,
but
they
don't
know
why
it's
important
to
have
it.
You
know
have
this
software
versus
some
other
software.
B
So
that's
why
you
know
the
other
thing
about
having
these
different
papers
is,
want
to
write
them
up
differently,
with
the
different
sort
of
set
of
things
going
associated
with
them
and
like
maybe
even
a
different
kind
of
language.
So,
like
you
know,
cleaning
up
a
lot
of
jargon
that
you
might
use
in
a
computer
science
paper.
B
So
where
we
talk
about,
you
know
very
machine
learning,
specific
terms
in
the
benchmarking
paper.
That's
excellent
in
the
General
Paper,
probably
not
because
people
don't
know
what
certain
pieces
of
jargon
are,
and
you
know
you've
read
through
those
papers
that
I
sent
out
on
C
elegans
and
you
run
into
that
a
lot.
You
get
like
a
lot
of
terms
that
what
does
this
mean?
So
that's,
that's
always
the
problem.
We
have
with
reading
discipline,
specific
papers
and
things
like
that.
C
We
have
an
important
product
right,
I'm
in
C
sharper.
It
was
not
hosted
on
different
platform
or
something
like
that.
We
can
use
it
just
giving
an
image
and
extracting
the
features
we
want,
but
right
now
the
development
platform
right,
which
is
there
with
us
just
a
minute.
Yes,
I'm,
sharing
my
screen.
Okay,.
F
No
I'm
fine
I'm
just
that
Clear
Lake,
okay,.
A
A
B
A
B
I
think
I've
seen
this
before
this
is
for
like
live
deployment
right.
C
Oh
sorry,
sorry,
sorry,
sorry,
right
now
we
have
the
end
to
end
product
end-to-end
products
which
is
deployed
in
render
can't
we
write
a
paper
on
this.
Like
we
have
built
this
software,
we
trained
the
model
on
imagenet,
reads
and
right
now
it
is
giving
a
good
accuracy.
There
are
three
different
type
of
models:
one
is
the
cell
segmenter
one
bread
segmenter
and
what
they
do.
The
whole
paper
will
be
about
this
and
actually
I
have
gone
through
all
papers
about
C
Indians.
C
There
is
no
good,
this
kind
of
software
which
it
hosts
on
platform
and
extract
the
features
we
want
so
I
was
hoping
like
I.
Have
this
thing
in
my
mind,
like
write
about
a
paper
about
whatever
we
software
we
built
right
now,
devolin
is
not
working.
I
mean
there
are
some
capricated
libraries
I
will
update.
I
will
update
it
for
sure
and
make
it
working
by
the
end
of
the
publishing
of
paper,
but
can't
we
write
a
paper
about
this
because
see
whatever
we
have
three
different
type
of
models.
C
There
is
a
four
different
type
of
models.
Actually,
cams
is
also
there.
It
is
just
not
hosted
on
the
platform.
I
can
do
that.
I
can
convert
into
a
index
type
and
I
can
host
it
on
the
development
platform
here
and
anyone
can
use
and
I
have
gone
through.
The
whole
I
have
done
some
literature
survey
on
the
internet,
but
I
find
something
like
like.
If
we
run,
we
are
getting
sorry,
no.
C
C
What
we
call
explaining
about
the
software
what
we
did-
and
there
is
no
software
to
compare-
also
there
is
C
SharePoint,
but
we
need
to
run
it
in
our
local
machine
and
show
it
like
all
these
things.
But
this
is
a
software
which
is
hosted
on
the
web
and
I
told
to
Minot
about
this
thing
and
he
was
like
yeah.
C
If
there
is
no
software
like
this,
which
is
boost
on
a
platform,
why
can't
we
write-
and
this
would
be
our
first
paper
explaining
and
segmenting
cells
on
web
and
showing
it
as
an
example
extracting
a
CSV
file.
I
will
add
this
feature.
Also
extracting
these
points
in
a
CSV
file,
just
download
the
CSV
file
and
internal
Network
and
try
to
make
the
graph
what
we
need.
C
C
There
is
no
I
mean
something
like
this,
which
is
hosted
on
the
platform,
so
my
main
Moto
would
be
like
writing
a
paper
on
how
we
trained
the
model
on
which
ways
we
have
time
what
methods
we
we
did
and
how
is
the
accuracy
with
the
ground
rules
and
comparingly
and
hosting
it
on
the
software
platform
that
someone
can
use
it
as
pink
and
extracting
the
CSV
file
according
to
different
time
frames,
yeah,
so
I
hope
we
can
do
that
right.
I
mean
writing
a
paper
about
devil
and
only
whatever.
We
have
right
now.
B
Oh
yeah,
that
would
be
good
and,
like
I,
said,
that's
something
that
we
could
decide
whether
we
wanna
at
what
level
we
want
to
write
that
so
we
could
well.
We
could
actually
write
it
at
two
levels.
B
We
could
have
something
like
that
in
the
overview
paper
kind
of
go
into
detail
about
it,
but
not
really,
like
you
know,
with
a
lot
of
the
technical
details,
but
that
might
be
for
like
a
paper
where
we
Benchmark
things
or
like
kind
of
really
dig
into
the
details
of
technical
details
of
it,
because
you
know
we
have
people
who
you
might
want
to
use
the
software
as
like,
and
you
know
an
online
tool
and
they
might
want
to
you
know
they
might
have
a
familiarity
with
hugging
face.
B
C
Yeah
mainly
by
all
this,
because
they
don't
know
about
the
machine,
learning
and
deep
learning
things
biology
and
all
this
computation
of
cells,
but
yeah.
If
we
provide
something
like
this,
they
can
use
it
right
in
a
paper
that
could
I
have
this
thing
only
in
my
mind,
right
now,
I
was
thinking
about
this
and
I'm,
going
through
my
next
code,
what
he
did
Etc
yeah.
F
C
B
A
B
For
other
people's
benefit
to
so
this
is
the
paper
I
was
talking
about,
so
this
is
something
that
was
written
largely
from
a
lot
of
the
documentation
and
put
together
in
a
you
know,
a
basic
overview,
so
there's
a
short
abstract,
the
summary
accessibility
statement
of
need,
so
the
statement
in
need
is
usually
like.
You
know,
that's
something
that
you
can
say
okay.
B
This
is
why
you
would
need
this,
and
actually
the
point
you
made
about
like
this
being
posted
on
the
web
or
you
know
being
able
to
use
it
as
something
hosted
on
the
web
is
something
you
should
add
to
this
section,
but
it
basically
talks
about.
Why
would
you
do
this?
Accessibility
just
means,
like
you
know,
where
do
you
find
it,
which
is
probably
should
be
up
at
the
top?
This
is
kind
of
a
summary,
so
this
the
abstract,
is
kind
of
a
very
short
summary
of
what
happened.
The
summaries.
B
Just
what
you
know
everything
that's
in
the
software
will
update
that
with
some
of
the
features.
Actually,
we
should
probably
break
it
down
into
the
different
toolboxes
as
well.
They
didn't
really
do
that
here,
but
that's
because
that's
something
that's
been
worked
on
this
Summer.
That
should
be
at
the
top.
In
the
summary,
then
you
have
technical
details.
B
This
needs
to
be
filled
out
a
lot
more
segmenting
and
embryo,
and
that's
something
that
you
know
we
can
actually
contribute
more
from,
like
you
know,
other
sources
and
and
other
things
you're
thinking
about
this
was
just
very
a
very
basic
description
and
then
you
know
maybe
like
a
little
demo
and
then
other
things
like
hyper
parameter,
optimization,
lineage
prediction,
meta,
feature,
detection
and
future
directions
which
obviously
have
to
be
changed.
B
We've
changed
our
sort
of
road
map
now
so
we'll
we'll
change
that
accordingly,
but
I
mean
this,
is
you
know
just
kind
of
the
structure
of
a
very
general
paper
where
you-
and
you
know
there
is
a
little
bit
of
jargon
in
here,
but
it's
really
kind
of
at
a
high
level
and
then
a
very
deep
level
would
be
like
a
lot
of
the
benchmarking.
A
lot
of
the
you
know,
other
things
to
consider.
B
B
Like
you
know,
we've
talked
about
certain
issues
in
the
code
and
things
like
that,
but
it
may
be
useful
to
kind
of
go
through
some
of
those
issues
and
I'll
also
be
talking
about
like
maintenance
issues
too,
because
you
know
people
come
in,
they
want
to
know
like
what
are
what
do
we
usually
have
to
do
to
keep
things
running.
You
know
so,
like
obviously,
python
libraries
always
change
and
get
deprecated,
and-
and
it's
just
kind
of
hard
to
keep
up
with
that.
So.
F
C
The
requirements.txt
file
now
according
to
what
we
need
that
we
can
do
it
and
right
now
the
update
would
be
on
the
paper
you
wrote
in
research.
Gate
was
like
just
you
know,
explain
the
more
technical
details
about
resonate
18
structure,
how
we
implement
it.
There
are
so
many
things
that
is
I
have
read
that
paper.
Actually
it
was
their
own
research
field
typing.
So
I
read
that
paper.
There
are
so
many
things
not
mentioned
right
now,
because
we
have
updates.
We
have
a
host
our
things
in
hugging
phase.
C
Also
again,
please,
and
we
have
the
model
right
now,
which
is
working
and
more
technical
details
about
what
resonated
structure
data
pre-processing,
because
whatever
data
we
have
there
is
something
called
noise
in
them.
We
added
some
different.
We
removed
some
noises,
we
done
some
image
processing
and
how
we
did
the
data
for
pre-positing
and
yeah.
Imagine
it
creates
what
we
used
and
IOU
score
Etc.
C
C
A
B
C
Yeah,
well,
if
you
don't
mind,
I
will
share
a
dog
with
you.
Minor
can
grow
this
photo
and
yeah
I'll
start
writing
out
the
paper
every
weekend.
You
just
review
it
and
whatever
changes
made,
if
I
write
and
Abstract
from
that
people
just
read
that
abstract,
whatever
changes,
I
need
to
add
Etc,
you
just
mention
it
there
or
comment
on
that,
like
you
did
for
Microsoft
proposal.
Okay,
that
would
be
fine
and
I
will
write.
Start
writing
the
paper.
After
reading
the
documents
which
you
have
shared,
that.
C
B
You,
the
other
thing
yeah
I
wanted
to
talk
about
with
that
is
that
we
have
I
have
a
library
of
papers.
This
is
stuff
that
we
had
like
across
the
meetings
that
maybe
in
the
last
three
years
or
so
that
we've
talked
about
with
respect
to
Devo,
learn
and
some
of
the
things
that
other
people
have
done.
So
you
know
they're,
like
this
paper,
for
example,
is
volume
segregation
programming
in
a
nematodes
early
embryogenesis.
This
is
from
physical
review
e.
B
This
is
a
you
know,
some
authors
and
they
they've
done
this
work
on.
This
is
kind
of
I,
think
more
biologically
oriented
paper,
but
they've
been
doing
these
things
with
you
know,
segmenting
cells
and
and
doing
other
types
of
fate
specification
and
looking
at
Auto
exploration.
B
But
it's
it's
definitely
worth
at
least
looking
over
and
we'll
get
into
some
of
these
papers
later
then,
there's
this
more
accurate
3D
atlas
of
C
elegans
neurons.
This
is
a
paper
where
they
actually
take
neurons.
They
segment
the
neurons
out
of
C
elegans,
set
of
CEO
images
of
C,
elegans,
volumetric
images
and
they're
able
to
build
an
atlas.
B
So
the
atlas
building
is
you
know,
that's
not
really
what
we're
doing,
but
a
lot
of
the
work
that
people
have
done
with
segmenting
things,
they're
building
things
like
atlases,
so
they're
putting
them
some
into
some
other
thing
that
they're
you
know
displaying
on
the
web.
So
a
lot
of
underlying
you
know
work
that
needs
to
go
into
say.
B
Like
an
analysis
at
at
heart
really
involves
like
segmenting
cells
and
doing
things
like
that,
so
there's
this
other
paper
about
establishing
a
more
of
a
morphological
atlas
of
the
embryo,
using
deep
learning
based
4D
segmentation.
This
one
is
probably
the
closest
more
most
relevant
one
to
what
we're
doing
the
sea
shaper.
So
this
is
what
social
moth
was
talking
about
with
C
shaper.
This
actually
is
the
4D
is
the
three
dimensions
of
space
plus
time
and
they're
working
on
this
problem
and
then
there's
a
lineage
result
molecular
Atlas.
B
So
this
is
where
you
have
an
atlas
of
what
they
call
molecular
data,
which
is
like
the
transcriptome.
So
it's
like
when
genes
are
regulated
by
by
different
processes
they
produce
transcripts.
Those
transcripts
can
be
measured
with
rna-seq,
which
we've
mentioned
in
the
meetings,
and
this
is
just
basically
a
measurement
of
gene,
expression
and
you'll.
You
know
it'll
vary
by
cell
by
cell
and
you
can
actually
build.
B
B
This
is
just
kind
of
a
paper
about
this
other
paper,
so
this
is
a
kind
of
a
article
summary
there's
fast,
deep
learning,
correspondence
for
neuron,
tracking
and
identification
since
using
synthetic
training.
B
So
this
is,
you
know,
kind
of
some
of
these
methods
that
they're
using
fast,
deep
learning,
correspondence
or
fdlc
they're
doing
this
in
C,
elegans
I
think
this
is
like
a
technique:
they're
working
on
they're,
just
using
C
elegans,
because
that's
a
simple
system,
but
this
is
the
kind
of
thing
that
you
know
what
we're
doing
with
a
lot
of
the
deep
learning
here
is
we're
trying
to
improve
upon
some
of
the
previous
techniques
and,
of
course,
the
original
technique
was
to
hand
draw
the
cells
and
the
neurons
and
the
C
elegans.
B
So
this
is
all
hand
drawn
at
one
time
and
then
you
know,
there's
different
recognition
approaches.
One
can
use,
and
so
they're
doing
this
fast,
deep
learning
correspondence
and
it's
a
basically
a
Transformer
and
they're
training
it
on
synthetic
data,
which
means
they're
making
data
up
based
on
the
distributions
you
find
in
C
elegans,
and
then
it
predicts
on
actual
samples
of
data.
So
this
is
something
that
you
know.
This
is
the
way
they're
doing
this.
B
I
don't
know
if
this
is
useful,
but
it
does
give
you
some
I
think
some
guidance
in
terms
of
what
people
other
people
are
doing.
This
is
the
same
paper,
I
think
and
then
there's
the
one
we
talked
about
a
couple
points
ago
and
then
this
is
a
book
I
think
but
I
think
there's
a
chapter
in
here
where
we
picked
for
C
elegans
for
4D
analysis
and
oh
it's.
This
paper
here,
chapter
two
automatic
automated
acquisition
of
cell
lineage
through
4D
microscopy
and
Analysis
of
early
C,
elegans
embryogenesis.
B
So
this
is
the
I
think
the
paper
that's
relevant
here-
and
this
is
you
know
where
they're
doing
this
basically
they're
constructing
the
cell
lineage
they're,
using
the
data
to
do
this,
and
this
is
a
problem
because
we
we
have
the
labels
for
the
cells
but
like
attaching
those
labels
to
the
cells
and
in
images
that
are
collected
at
different
points
in
time
and
then
reconstructing
all
that
is
kind
of
a
hard
problem
for
someone
who's
just
kind
of
starting
out
or
you
know,
doesn't
know
a
lot
about
C,
elegans
or
whatever.
A
B
There's
there's
some-
maybe
some
pointers
in
this
article
here
chapter
two
in
this
book,
but
and
then
there's
this
ZIP
file
of
spatial
distributions
for
deep
learning.
So
this
is
actually
another
I,
don't
know
how
many
papers
are
in
this,
but
that
that's
another
thing
you
might
want
to
check
out.
So
there
are
a
bunch
of
papers
in
there
that
you'll
find
useful.
B
So
it
looks
like
Lucas
has
a
question.
He
has
his
hand
raised.
E
But
yeah
I
was
just
wondering
if
you
could
share
the
link
like
the
Google
drive,
folder
yeah.
B
Yeah
I'll
put
it
in
the
slack
for
people
and
then
I'll
send
it
directly
to
you
Lucas
yeah,
so
yeah
I
recommend
people.
People
want
to
read
it
Beyond,
like
you,
know,
sushmath
and
geothi,
because
there
are
a
lot
of
good
papers
in
there
about
some
of
these
problems
so
yeah.
So
Lucas
is
here
in
size
here.
So
Lucas
has
been
working
on
he's
interested
in
contributing
to
some
papers
and
he's
been
working
on
some
ideas
and
he
so
do
you
want
to
give
an
introduction
Lucas.
E
Yeah,
so
my
name
is
Lucas
I'm
an
undergrad
student
in
Canada
and
I
joined
I
actually
knew
there's
a
lab
called
orthogonal
lab
and
it
was
Mutual
like
I
knew
it
from
last
year,
but
like
I
joined
the
lab
this
year
and
I
was
I
just
joined
the
lab
to
look
for
opportunities
like
research
opportunities
and
I'm,
also,
like
I,
think
I'll
be
able
to
present,
like
it's
like
a
presentation
on
biophysics
the
biophysics
of
the
red
light,
which
I
thought
it's
interesting,
not
necessarily
I.
E
Don't
think
it
would
be
an
interesting
topic
of
research
for
like
because
you
need
some
facilities
for
doing
so.
For
example,
what
I
was
thinking
as
an
extension
to
like
the
biophysics
of
the
pet
flight
is
like
setting
up
an
experiment
to
find
like
a
you
know,
in
the
lift
Force
like
drag
force
and.
A
E
Equations,
like,
like
the
lift
Force,
if
you
search
it
up,
the
equation
is
CL,
which
is
a
constant
times.
One-Half
row
V
squared
S
V
squared,
is
the
velocity.
S
is
the
surface
area
of
the
ring,
and
this
CL,
which
is
technically
like
a
constant
number,
which
varies
from
different
angle
of
attacks
and
different.
E
You
know
different
different
density,
Etc
I
I
was
wondering
like
this
again
is
for
the
people
who,
like
our
technology,
but
again,
you
could
set
up
an
experiment
to
let's
say,
find
the
CL,
the
certain
constant
number
for,
like
a
certain
Birds
PC
and,
like
let's
say,
a
different
angle
of
attacks
and
then
publish
it
online.
E
So
the
people
for
like
the
orphanologists
who
are
interested
in
this
actual
like
stuff
like
the
serial
number,
could
use
it
and
yeah,
but
it's
not
necessarily
something
that
I've
found
that
we
could
do
it.
But
again
it's
just
a
presentation
that
I'm
gonna
be
doing
tomorrow
and
and
then
like
as
I,
said
I'm
generally
interested
in
research
opportunities,
and
if
someone
has
anything
to
present
or
like
has
any
idea,
I'd
be
interested
to
contribute.
If
I
see
I
could
yeah.
B
Yeah,
that's
great,
thank
you
Lucas
for
that
update,
so
yeah
they're
always
opportunities
to
do
things
in
the
in
the
group.
Here
you
know
we
go
over
a
lot
of
things
during
a
meeting
and
so
don't
feel
overwhelmed,
but
also,
if
you,
if
you
want
to
reach
out
about
something
that
you
know,
is
interesting
to
you
just
please
do
and
we'll
try
to
put
something
together.
B
B
Something
we
can
do
is
like
you
know,
you
know
an
easy
sort
of
research
agenda.
Sometimes
it
takes
like
a
lot
of
effort
to
go
and
find
the
right
resources
and
things,
but
you
know
we
have
to
kind
of
get
things
off
the
ground.
We
have
to
discuss
the
how
to
do
them
and
we
have
to
give
you
know.
Just
kind
of
the
idea
here
is
like
present
ideas
and
then
work
out,
work
them
out
to
some
extent,
and
sometimes
we
get
things
completed.
B
Sometimes
we
don't,
but
it's
always
you
know,
don't
don't
hesitate
to
talk
about
what
you're
interested
in
yeah.
A
B
So
then,
please,
please
yeah
get
in
contact
sigh.
How
are
you
seeing
you
in
the
meetings
before.
C
B
And
yeah
the
same,
the
same
thing:
Ultra:
are
you
in
our
slack,
which
is
our
communication
channel
for
evil
worm?
If
you're
not,
please
put
your
email
in
the
chat
so
I
can
it's
not.
You
know
if.
B
Okay,
yeah,
that's
great!
So
that's
great!
So
again
the
same
thing
holds
true
that
I
told
Lucas
if
you're
interested
in
some
topic
and
want
to
follow
up
what
let
us
know
so.
Yeah
I
see
that
dick
put
in
the
chat
here
that
he
was
Dr.
He
dreamed
all
night
about
mapping
differentiation,
trees
of
the
genome
and
then,
as
an
aside,
anyone
study
graph
Theory,
which
is
not
networks
but
I,
don't
know
if
you
want
to
talk
about
that
more
dick.
A
G
Okay
Bradley:
this
is
for
you
to
do
I
found
one
tree,
but
it
doesn't
obviously
doesn't
show
all
800
or
900
cells
right.
A
B
Okay,
this
this,
oh,
the
when
the
drafty
sent
me.
G
A
G
B
G
Okay,
yeah,
that's
sort
of
obvious,
but
okay,
now
one
thought
I
had
is
the
metabolism
of
organisms
have
been
worked
out
very
well.
G
G
G
G
G
G
Then
you
may
have
the
right
thing
and
I
looked
it
up,
and
the
number
of
cell
types
is
far
lower
than
the
number
of
repeats
of
repetitive
DNA.
A
G
Typically,
go
into
the
hundreds
of
thousands,
you
know
no
organisms
made
of
hundreds
of
thousands
of
different
kinds
of
cells,
so
yeah,
so
it
might
be
able
to
distinguish
it
that
way:
okay,
yeah
so
I'm
suggesting.
Basically
we
might
be
able
to
find
some
sort
of
motif
by
first
looking
at
metabolism
and
then
see
if
we
can
find
a
motif
that
corresponds
in
numbers
will
remember:
cell
types.
A
A
F
G
Okay,
because
it
sort
of
unconceivable,
if
you
can
find
it.
A
B
G
G
B
Yeah
the
whole
thing
about
cell
types
in
like
mammals
and
these
sort
of
regulative
development
contexts.
More
generally,
is
that
you
don't,
like
you,
have
a
lot
of
precursor
cells.
You
have
a
lot
of
things
that
are
kind
of
type
like
so,
like
you
know,
they're
not
quite.
G
A
G
What
you
might
call
cryptic
cell
types
where
you
can't
distinguish
them,
because
you
have
no
measurement
now,
but
they
might
be
different.
Yes,
yeah,
that's
right!
That's
that
concept
came
out
of
the
evolution
of
whole
animals.
There
are
cryptic
species,
the
Delta.
Is
they
don't
interbreed,
but
they're
I
think.
As
a
result,
they
don't
interbreed,
but
you
can't
tell
them
apart.
G
A
B
Okay,
so
yeah
I
had
some
things
actually
to
show
about
this.
If
I
I
want
to
share
my
screen
now
so.
G
B
So
that's
thank
you
for
that.
That's
good
I
saw
a
couple
of
versions.
Come
across
the
email,
so
I
have
I.
Have
those
in
my
possession
so
I'll
take
a
look
at
them.
I've
been
looking
them
over
a
little
bit,
but
so
oh
yeah,
we
had
a
question.
Was
that.
B
So
we
had
a
question
last
week
about
like
different
types
of
wait:
how
cell
types
of
regulated
or
switching
between
cell
types
and
so
I
put
together
some
resources
on
this.
So
of
course,
there's
this
literature
on
epigenetic
switches,
and
things
like
that.
So
I
think
this
is
where
I
put
them:
okay,
yeah!
So
there's
all
this
literature
and
like
these
epigenetic
switches.
So
you
know
that's
really
what
I.
B
So
there's
this
these
set
of
epigenetics,
which
is
that
exist
in
the
genome,
oftentimes
they're
associated
with
certain
cell
States,
and
so
you
have
this.
You
have
these
cpg
islands,
these
cytosine,
the
guanine
Transitions
and
there's
a
methyl
group
that
sits
on
top
of
that
and
it
forms
this
epigenetic
switch
where
it
switches.
You
know
what
DNA
is
read
during
transcription,
so
that
you
have
different.
That's
basically
the
controls
different
cell
types.
B
So
when
you
have
this
state,
the
switch
can
be
in
different
states
and
it
result
It's
associated
it's
not
it's
not.
They
haven't
necessarily
demonstrated
this,
but
it's
always
associated
with
different
cell
types
and
critically
there's
a
by
stability
of
the
switch.
So
the
switch
can
be
bystable.
B
It
can
switch
from
like
say
a
stem
selfie
to
a
differentiated
selfie,
and
so
this
is
one
of
the
mechanisms
that
they
talk
about
these
epigenetic
changes,
and
so
these
are
very
important,
at
least
in
the
literature
where
they
talk
about
some
of
these
when,
when
they
look
at
like
say,
for
example,
cells
that
have
been
reprogrammed
where
they
look
at
the
the
Fate
restriction
of
stem
cells,
they
often
talk
about
they
often
assay
for
this,
so
they.
B
This
is
a
review
article
on
epigenetics
stem
cells
and
epithelial
sulfate,
and
it
kind
of
goes
through
some
of
this.
You
know
this
work,
you
know
where
you
have
there.
You
know
you
can
go
through
some
of
this,
what
they
call
the
histone
code,
which
is
a
lot
of
the
stuff
that
has
to
do
with
some
of
these
epigenetic
Transformations
this.
These
epigenetic
Transformations
play
essential
roles
in
the
establishment
of
transcriptional
programs
accompanying
cell
differentiation,
and
so
they
talk
about
that
quite
a
bit
here.
B
So
that's
just
like
you
know
in
different
cell
Fates,
say
in
in
they
will
I
think
this
is
mainly
in
mammals,
I
think
in
C
elegans.
They
have
I,
don't
know
if
they've
established
what
the
actual
markers
are,
but
this
is
another
article
epigenetic
regulation
of
stem
cell
feet.
This
is
on
where
they
kind
of
go
over
some
of
the
potential
of
looking
at
this
and
they're
talking
about
here.
The
research
that's
been
done
so
far
is
only
the
tip
of
the
iceberg
when
it
comes
to
writing
an
epigenetic
instruction
manual.
B
So
this
is
like
something
that
people
are
kind
of
working
towards.
Like
saying
and
this
in
this
literature,
they
have
a
tendency
to
think
that
it's
all
about
this
epigenetic
aspect,
so
this
is
kind
of
one
of
these
things
where
you
know,
if
you
don't
include
this
in
the
paper,
they
get
a
they
get
antsy.
If
you
talk
about
like
you,
know,
selfie
but
anyways,
sometimes
people
can
fixate
on
certain
mechanisms,
but
this
is
important
for
cell
commitment
and
differentiation.
B
So
this
is
another
paper,
so
I
I
just
brought
up
a
couple
over
review
papers
on
this
and
then,
of
course,
there's
this
epigenetic
memory
aspect.
So
there's
this
aspect
of
of
you
know
transitioning
through
cell
types.
So
usually,
we
have
like
a
stem
cell
and
it's
pluripotent
and
it
can
maybe
differentiate
into
several
different
cell
types
and
then
those
that
so
the
Fate
is
restricted
to
those
cell
types
and
then
you
can
differentiate
it
further
into
other
cell
types.
B
B
B
So
this
talks
about
embryonic
and
trophoblast
stem
cells
reflect
the
first
irrevocable
selfie
decision
and
development
and
is
reinforced
by
distinct,
epigenetic
lineage
barriers.
So
again,
these
epigenetic
marks
the
the
hypothesize
are
the
things
that
are
stopping
it
or
can
can
push
it
into
a
new
fate.
Nonetheless,
embryonic
stem
cells
can
seemingly,
where
TS
like
character,
trophoblast
cell
stem
cell,
by
characteristics
upon
manipulation
of
lineage,
determining
transcription
factors,
reactivation
of
the
extracellular
signal,
regulated
kinase,
one
slash
two
or
irk12
pathway.
B
So
there's
this
pathway
that
is
involved
in
this,
so
they've
interrogated
the
progression
of
reprogramming
and
ESL
models
with
regulatable,
opt4
and
cdx2
transgenes
that
those
are
the
genes
that
are
involved
in
the
stemness
OR,
maintaining
the
stem
cell
state
and
so
they're
able
to
initiate
differentiation,
but
lineage
conversion
remains
incomplete
in
all
models,
underpinned
by
the
failure
to
demethylate
a
small
group
of
TS
cell
genes.
B
So
there's
this
whole
model
of,
like
you
know
how
these
things
get
regulated,
you
can
do
experiments
to
stop
things
from.
You
know,
differentiating
and
things
like
that,
and
then
there's
this
paper
and
molecular
Hallmarks
of
epigenetic
control,
transition,
States
and
sulfate
decisions
and
epigenetic
Landscapes.
So
this
links
this
to
waddington's
landscape,
where
they
kind
of
we've
seen
waddington's
landscape.
Where
you
have
this
landscape,
that's
like
a
tree
where
it
branches
off-
and
those
branches
of
course,
are
the
things
we're
trying
to
capture
with
different
cell
differentiation
and
they
kind
of
Link.
B
So
if
you
go
down
to
the
I,
don't
know
if
they
have
yeah,
they
have
an
image
of
a
landscape
here,
where
you
have
a
ball
rolling
down
a
hill
into
these
channels
and
the
channels
Branch
off,
so
that
you
have
these
State
changes,
I
guess,
and
so
it's
very
metaphoric.
This
is
supposed
to
be
a
gene
expression
network.
But
it's
not
really.
You
know
waddington's
time,
they
didn't
know
a
lot
about
that,
so
it
this
is.
B
So
it
can
attain
one
of
two
states
by
being
in
this
meta,
stable,
State
and
there's
this
feedback
loop
here
between
X
and
Y,
where
these
are
the
two
cell,
States
X
and
Y,
and
there's
this
feedback
that
keeps
it
in
this
in
the
stable
state.
But
there's
also
this
break
also
that
keeps
it
from
converting
to
another
cell
type
when
that
break
is
removed.
Of
course
you
can
go
to
the
new
cell
type
when
the
break
is
applied.
B
You
stay
in
the
same
cell
state,
so
there's
this
metastable
state
that
involves
both
feedback
and
in
this
break.
So
that's
that's
what
they're
talking
about
when
they
talk
about
a
metastable
state?
It
you
know,
keeps
it
in
this
sort
of
it
keeps
the
potentiation
there,
but
it's
also.
You
know
there
are
factors
that
keep
it
from
transitioning
to
another
cell
type.
Then
there's
this
Dynamic
model,
which
is
where
you
have
this
amount
of
stable
State
and
then
it
bifurcates
into
this
by
stable
state.
B
So
at
some
point
this
stable
cell
X
will
hit
this
bistable
switch,
meaning
that
it
can
either
remain
X
or
go
to
Y
and
so
or
it's
a
stem
cell
that
bifurcates
into
X
or
Y.
So
those
are
the
two
models
they
use
here
and
they're.
Thinking
about
this
in
terms
of
waddington's
landscape.
So.
G
B
A
E
Okay,
I
have
a
question,
so
you
see
on
the
buy,
stable,
State,
there's
a
continue,
dotted
line.
Yeah.
A
E
B
So
this
this
dotted
line
is
just
kind
of
like
if
you
see
that
up
at
the
top
and
see
there's
this
meta,
stable
State.
This
would
be
the
meta
stable,
State
continued
from
like
oh,
okay,
okay,
so.
B
So
the
monostable
is
like
where
you
have
a
single
cell:
it's
a
stem
cell
and
it
doesn't
changed
anything.
And
then
you
hit
the
switch
point
here
where
the
red
lines
diverge
and
it
can
either
become
one
state
or
another.
There's
this
regime
in
the
middle,
which
is
imaginary,
but
it's
basically
a
thing:
that's
keeping
these
on
either
side.
So
like
yeah,.
B
So
yeah
they're
they
do
mention,
like
you
know
they
talk
about
the
mechanisms
for
this
in
terms
of
like
the
cell
State
splitter
analogy,
they
talk
about
things
like
chemical
equilibrium
and
things
like
that.
But
no
one
knows
really
what
the
thing
is.
It's
keeping
these
things
other
than
like
an
epigenetic
marker,
which
of
course,
is
not
necessarily
I.
Don't
I
mean
I,
don't
think.
That's
all
that's
happening
because
it
seems
like
there's.
You
know,
there's
a
lot
more
in
the
cell.
That's
probably
happening,
but
that's
like
one
thing.
We
can
assay.
B
That's
a
quick
and
easy
so
like
this,
this
paper
talks
about
epigenetic
switching
is
a
strategy
for
quick
adaptation
while
attenuating
biochemical
noise,
and
so
this
talks
about
like
some
of
the
systems
level,
things
that
are
happening
and
these
epigenetic
switches
again
are
used
as
a
sort
of
the
stand-in
for
the
stability
or
the
ability
to
maintain
like
a
bi-stable
state
or
whatever,
and
then
they
talk
about
this
in
terms
of
environmental
noise.
B
So
if
you
have
a
fluctuating
environment
or
you
have
noise
in
the
presence
of
noise,
this
is
something
that
reduces
the
amount
of
noise.
So
if
you
have
this
fluctuating
environment,
you
know
it
can
potentially
push
cells
around
the
different
states.
B
If
they,
if
you
attenuate
that
noise,
you
have
a
better
chance
of
keeping
it
in
that
state.
The
noises,
of
course,
could
be
things
like
heat
or
you
know
other
types
of
things
like
osmotic
pressure.
Those
are
things
that
vary
over
time.
So
if
there's
a
lot
of
variance,
it
can
affect
the
cell
and
it's
state,
it's
it's
transcriptional
program,
and
then
it
can
push
it
into
a
state,
maybe
even
into
a
cancer
state.
B
B
This
as
a
marker,
this
is
probably
not
the
only
thing
that's
happening,
but
there
is
actually
a
signature
in
the
genome
for
where
these
epigenetics,
which
is
our
they
call
it
a
CP
cpg
transition
or
a
cytosine
to
guanine
transition,
and
they
have
these
cpg
islands
that
you
can,
that
that
are
in
I,
don't
know
if
they
have
a
cpg
Iowa
map
or
CL
against,
but
I
know
that
they
have
them.
B
They
have
characterized
them
in
other
genomes,
and
so
this
is
again
epigenetic
and
transcriptional
regulations,
Prime
selfie,
before
division
during
human
pluripotent
stem
cell
differentiation.
This
is
again
like
priming
selfie,
so
people
are
always
looking
for
how
selfique
it's
primed
or
you
know
what
the
factors
are,
that
kind
of
started
down
that
path,
and
they
talk
about
this.
B
This
is
like
the
method
that
they're
using
people
have
developed
all
sorts
of
methods
for
like
looking
at
these
things
and
then
General
thing
is
like
you
know:
you'll
you'll,
maybe
even
try
to
induce
it
with
some
factor
and
then
look
at
the
transcriptional
program
before
and
after
and
if
you
know,
if
you
get
you
can
find
like
signatures
of
of
that,
not
only
the
cell
type
but
of
differentiation
they
have
people
haven't
really
looked
at
differentiation
in
that
way,
but
like
they'll,
look
for
different
markers
of
a
cell
type
and
they'll
say
you
know
this.
B
This
cell
expresses
these
these
genes
and
they're
upregulated.
So
that
must
mean
that
it's
it's
a,
maybe
a
muscle
precursor
or
it's
a
muscle
cell,
because
it's
expressing
these
genes-
and
you
know
you
can
look
at
the
morphology
of
the
cell,
where
the
what
the
cell
looks
like
under
a
microscope.
But
it's
still
kind
of
hard
to
you
know
it's.
It's
still
kind
of
one
of
these
things
where
you're
just
basically
saying
that
it's
a
cell
type
because
of
these
things
and
has
these
these
genes
up
regulated.
B
So
it's
really
kind
of
a
you
know:
I
don't
think
people
have,
and
then
this
is
probably
not
relevant,
but
yeah
I.
Think
that's
that's
like
when
you
mentioned
that
last
week,
I
had
to
go
and
look
in
the
literature
and
see
what
the
state
of
the
art
was
in
terms
of
what
people
are
thinking
about.
So
I
I
don't
know
I.
B
You
know
I
think
like
the
idea
about
getting
like
motifs,
that's
probably
where
the
opportunity
lies,
and
then
you
know
even
like
having
like
a
model
of
motifs
where
or
a
model
of
some
some
of
these
things
in
the
literature
and
incorporating
into
that
model.
So
you
can
say
well
you
know
we
have.
We
maybe
can
look
towards.
B
Maybe
the
way
genes
are
organized
on
a
chromosome
or
you
know,
different
patterns
of
expression,
different
switches
that
might
exist
in
the
genome
and
kind
of
making
those
kind
of
predictions
for
what
might
be
controlling
differentiation,
because
there's,
there's
cell
type,
things
that
identify
cell
types
and
things
that
identify
differentiation
and
it
seems
like
the
chromatin
markers-
are
sort
of
identify
differentiation.
B
But
it's
not
clear
like
if
that's
in
response
to
something
else
or
if
it's
just
maintaining
cell
State,
it's
not
really
clear
so
yeah
the
literature
is
not
set
up
to
answer
this
problem.
I,
don't
think,
but
that's
that's
not
a
problem
necessarily
I
think
people
think
about
it
differently
in
in
a
lot
of
like
in
in
the
biomedical
context,
people
think
about
it
differently.
F
I've
asked
chat,
GPT
a
number
of
things
and
I
always
ask
for
references
and
that
helps
with
a
reference
search.
Sometimes
it'll.
Take
you
off
into
a
area
where
well,
you
need
to
go
but
didn't
realize
it
was
there
yeah.
B
D
B
Don't
think
so,
but
but
that's
you
know,
that's
a
skill.
We
can
I
mean
I.
I
know
something
about
bioinformatics
I.
Guess.
F
B
B
Yeah,
it's
not
you
know.
Genome
working
with
genomes
is
not
necessarily
easy.
That's
the
thing
I
know
that
yeah
well,
it's
like
getting
like
raising
like
sort
of
making
conclusions
from
it.
I
guess
is
the
problem
because
it's
like
you
have
the
sequence,
but
then
what
does
that
mean?
We
have
annotations
they're
not
like
well
and
C.
Elegans
are
more
specific
than
say
like
in
humans,
but
they're
still,
you
know.
A
B
So
yeah
well
we'll
revisit
that
we'll
keep
working
on
that
final
thing,
I
want
to
talk
about
was
that
this
week
we
finished
a
newsletter
for
open
one.
So
I
wanted
to
go
over
that.
So
we're,
of
course,
part
of
the
open
worm
foundation,
and
this
is
the
open,
warm
news
for
this
month,
so
we've
we've
sometimes
put
out-
or
it's
actually
pore-
Gleason
who's,
one
of
the
senior
contributors
at
open
worm.
B
He
put
together
this
he's
putting
together
these
newsletters
and
I've
I've,
put
in
a
part
on
Diva
worm
and
everything.
So
so
we
have
these,
and
this
is
so
for
June
2023
looks
like
there
was
a
poster
presented
at
the
C
elegans
Conference.
So
the
c
elegans
conference
is
a
conference
on
the
people
who
are
interested
in
using
C
elegans
as
a
model
organism
for
research
for
looking
at
things
like
disease
and
for
looking
at
Chronic,
comics
and
and
basic
research.
B
They
get
together
at
the
c
elegans
meeting
every
two
years,
and
so
it's
really
good
because
it's
like
all
focused
on
c
elegans,
just
about
any
topic
you
can
think
of
and
so
that
it's
posted
in
Glasgow
Scotland
every
two
years
used
to
be
in
Los
Angeles,
but
they
moved
it,
and
so
they,
basically,
this
is
the
poster
we
have
I,
don't
know
if
this
is
gonna
open,
I
think
it
will.
B
So
this
is
the
poster
we
have.
The
introduction
where
we
talk
about,
like
you
know,
open
worm
is
an
open
science
initiative
where
we
have
open
data
code
and
people
can
run
open
worm
as
a
simulation.
They
can,
you
know,
use
the
data
things
like
that,
so
we
get
the
data
from
different
sources.
We've
made
data
available
from
different
labs
and
we've
talked
about
having
an
internship
with
people
from
different
biological
Labs
learn
how
to
build
open
data
sets.
But
those
are
things
that
you
know
all
involving
the
open
worm.
Foundation.
B
Then
you
know
there
was
this
one.
The
first
study
that
really
kind
of
came
out
of
open
worm
was
the
accurate
simulation
of
locomotion.
So
we
have
these
models
of
locomotion.
We
know
that
you
know
C
elegans
has
these
sort
of
stereotypic
movements
that
it
does
and
we
can
characterize
them,
link
that
to
the
electrophysiology
link
that
to
the
cellular
circuits
and
the
connectome
and
the
muscles.
So
there
are
these
different
Gates
that
we
can
see
under
a
microscope.
We
can
model
them
and
we
have
these
things
that
are
generating
those
movements.
B
A
B
A
sort
of
a
virtual
machine
for
your
computer,
you
can
set
it
up
on
your
computer
and
you
can
set
up
this
Docker
container
that
we
have
where
all
the
simulations
of
open
warmer
in
there,
and
so
you
open
it
up.
You
run
it
it'll
run
through
the
simulations
one
at
a
time.
There's
the
of
course
The
Locomotion
simulation
there's.
So
it's
like
cybernetic,
which
is
here.
B
You
have
other
types
of
simulations
of
the
connectome
like
neuro
like
c302
and
some
other
things,
and
then
you
know
you
can
run
those
in
sequence
and
see
kind
of
the
worm
and
pieces,
I
guess
or
in
different
subsystems.
It's
really
nice,
it's
being
still
being
developed,
and
so
there
are
a
lot
of
other
things
going
on.
There's
a
Hodgkin
Huxley
tutorial
that
happened.
B
This
is
a
Google
summer
code
project
that
porrig
was
a
mentor
on
and
this
person
Rahul
sunkar
developed
this
tutorial
for
a
hodgkin-huxley
model,
which
is
a
model
of
of
ion
channels
in
C,
elegans
and
C.
Elegans
doesn't
have
the
same
set
of
ion
channels.
We
do,
but
they
they
don't
fire
Action
potentials.
B
The
way
we
do,
but
there's
still
a
lot
of
things
to
learn
with
the
Hodgkin
Oxley
model
from
for
C,
elegans,
connectomes
and
behavior,
there's
also
neural
pal,
which
is
this
tool
that
looks
at
like
the
gene
expression
in
brains
in
in
worm,
brains,
I,
guess,
there's
the
simulation
stack.
This
is
Diva
worm
and
Divo
learned
I
just
put
together
a
graphic
of
a
different
thing:
the
different
areas
of
research.
We
do.
We
have
mathematical,
modeling
cell
segmentation
and
secondary
data,
and
you.
B
Is
kind
of
a
graphic,
it's
very
cramped
area
to
do
this
in,
but
you
know
we
just
have
our
little
feature
here.
We
have.
We
have
these
student
ships
where
we
get
students
to
learn.
You
know
to
do
something
in
the
foundation,
so
it
involves
a
small
stipend
and,
if
you're
interested
the
application
process
is
at
this
link.
Openwarm.Org
studentships-
and
it's
just
like
it's
not
it
doesn't
pay
very
much,
but
you
get
to
work
on.
You
know
something
in
the
open
arm.
Foundation
try
to
improve
it.
B
It's
kind
of
like
Google
summer
of
code,
the
stipend
isn't
as
big,
but
it's
still
kind
of
a
you
know.
A
nice
thing
to
put
on
your
CV
gets
you
into
a
project
yeah,
so
I
mean
you
know.
We
have
worm
body
models,
we
have
electrophysiological
recordings
for
different
cells,
so
in
C
elegans
all
the
cells,
including
the
connectum
cells,
have
an
identity,
so
those
are
all
like
identifiable,
and
then
we
also
have
single
cell
data
for
a
lot
of
things.
So
this
is
electrophysiological
recordings
from
the
ash
neuron.
B
So
we
have
a
specific
neuron.
We
can
record
from
that
neuron.
We
know
the
connections,
the
other
neurons.
We
can
record
from
those
we
can
put
together
circuits
which
are
connection
connected
neurons,
and
then
we
can
map
that
some
movement
Behavior
say
which,
of
course,
people
have
done
in
different
papers
where
they
look
at
like
if
a
worm
is
moving
forward
and
then
backing
up
what
is
happening
in
the
in
that
little
part
of
the
connected
on
that
circuit,
and
so
we
can
actually
characterize
this
quite
well.
B
So
this
is
stuff
that
we're
working
on
with
other
labs
outside
of
openworm,
and
then
you
know
there
there's
future
work
that
we
had
there's
a
lot
of
stuff,
there's
modeling
things
in
neural
ml,
which
is
a
type
of
XML.
So
it's
like
a
markdown
like
or
a
markup
language.
For,
like
the
connectome,
you
know,
incorporation
has
sensory
feedback
between
worm
body,
environment
and
neural
network.
B
So
this
is
where
your
characterizing
sensory
feedbacks,
the
different
types
of
Imaging
like
calcium,
Imaging,
there's
some
new,
a
lot
of
new
techniques
out
there
for
collecting
data
and
that's
something
we
can
sort
of
embrace
as
a
community.
B
So
I
think
this
is
nicely
because
it
kind
of
gives
you
an
update
as
to
what's
going
on
in
the
project,
so
yeah
this
just
kind
of
talks
about
all
of
that
in
the
in
the
newsletter
and
then
you
know,
they're
well
he's
constantly
trying
to
funder
activities,
and
then
you
know
the
student
ships
are
more
successful
for
this
past
year.
So
this
last
year's
studentship
was
where
there
was
this.
B
We
have
this
openworm
browser,
which
is
a
virtual
browser
of
the
worm,
and
so
you
can
take
it's
blender.openerm.org
and
you
can
look
at
the
worm
from
different
perspectives.
You
can
like
remove
the
muscles
that
were
the
cuticle
and
you
can
look
at
the
nerves.
You
know
in
the
neurons
you
can
put
the
muscles
and
cuticle
on.
B
You
can
look
at
the
whole
worm
you
can
zoom
in
you
can
get
inside
the
worm
to
some
extent,
so
there
this
had
to
be
improved
because
it's
it's
kind
of
dated
now
so
tasting
wheel,
right,
who's,
I,
think
a
graduate
student
at
the
University
of
Wisconsin.
Now
he
worked
on
this
doing
this
in
blender,
so
this
is
nice.
This
is
this
model
is
open
source,
so
you
know
people
are
interested
in
working
on
it
more.
It's
it's
possible
to
create
a
fork
and
to
work
on
it.
E
Ship
is
only
I
mean,
could
you
apply
as
an
undergrad
student,
or
should
you
be
graduate
or
PhD.
B
I'm
not
sure
what
the
rules
are
on
it.
I
don't
see.
Why
not,
but
I
think
it's
a
larger
way
for
students,
though
it's
like
you
know,
because
it
really
does
teach
you
you
can
apply.
If
you
wanted
yeah
I,
don't
know,
I,
don't
know
what
the
conditions
are
but
definitely
check
it
out.
Yeah
Lucas,
I'm.
B
Well,
I
I
worked
with
porg
on
it
last
year,
as
meant
as
a
co-mentor
and
kind
of
I.
Don't
think
the
rules
are
hard
set
in
stone
on
that,
but
I'm
not
also
not
in
the
like
in
the
board
meetings.
I,
don't
know
what
what
they're
thinking
in
terms
of
but
yeah
I
would
I
would
definitely
follow
up
on
that
yeah.
So
then
you
know
we
have
all
sorts
of
different
types
of
data
that
are
available.
We
have
you
know.
This
is
another
thing
too.
The
people
are
interested
in
projects.
B
Now
it's
well
sometimes
a
little
challenging
to
work
with
the
data,
because
it's
it's
high
in
a
highly
specialized
format
and
you
have
to
figure
out
what
the
variables
are
and
how
to
use
them.
But
it
you
know
it's
definitely.
You
could
do
a
lot
of
projects
with
the
open
data.
That's
out
there
in
CL
again,
so
it's
possible
to
do
a
lot
of
really
interesting
things.
So
that's
that's
another
thing
and
then
the
simulation
stack,
which
is
running
this
Docker
container
and
that's
something
that
I
think
is
underutilized
by
people.
B
It's
it's
a
nice,
it's
hard
to
run
because
you
know
you
have
to
run
it
on
the
right
machine.
I've
been
trying
to
run
a
large
language
model
on
a
local
machine
recently
and
it's
hard
to
get
these
things
to
run
locally.
So
it's
not
not.
You
know
it's
not
something,
it's
easy
to
do,
but
it's
something
that
I
think.
If
you
have
the
opportunity
to
set
this
up
and
get
it
running,
it's
really
cool.
B
So
that's
all
I
wanted
to
talk
about
with
respect
to
that.
I
just
want
to
give
an
update
on
on
open
worm
and,
what's
going
on
there.
B
B
A
B
Yeah
yeah,
we
do
yeah,
we
do
a
lot
of
what
we've
done
in
the
past,
a
lot
of
sort
of
informatics,
stuff
genetic
you
know,
working
with
genome
data
is
kind
of
challenging
because
you,
like
I,
said
you
have
to
have
like.
Sometimes
it's
in
specific
formats.
You
have
to
like
get
it
out
and
map
a
lot
of
these
things
to
your
problem
and
so
just
grabbing
some
Gene
sequences
isn't
really
the
only
thing
you
need
to
do
it
has
to
be.
B
All
right
anything
else
before
we
go
today.
B
Thank
you.
Okay,
now,
I
want
to
talk
about
a
few
papers
on
constructive
connectomics,
and
so
this
is
going
to
go
over
some
sort
of
recent
developments
and
connectomics
with
respect
to
C
elegans
and
some
alternative
neural
systems
and
as
we'll
see,
this
has
implications
for
the
greater
Enterprise
of
connectomics.
B
I
have
three
papers,
I'll
start
with
cl
again,
so
the
first
one
is
this
Frontiers
and
systems
Neuroscience
papers.
This
is
from
2021,
as
paper
is
called
Computing
temporal
sequences
associated
with
Dynamic
patterns
on
the
CL
against
connecto
all
right.
So
let's
go
over
the
abstract
understanding
how
the
structural
connectivity
and
spatial
geometry
of
a
network
constrains
the
Dynamics
of
the
network.
It
is
able
to
support
an
active
and
open
area
of
research.
B
We
simulated
the
plausible
Dynamics
resulting
from
the
known
C,
elegans
connecto,
so
we
know
the
C
elegans
connect
them
both
in
the
number
of
cells,
the
identity
of
those
cells,
hence
the
function
of
their
cells
and
the
connections
between
them.
So
this
is
an
excellent
opportunity
to
map
neurons
to
behavior
and
in
a
lot
of
ways,
people
do
that
already.
But
you
know
these
are.
B
You
have
sort
of
the
precursors
of
this
connectome,
so
we
have
the
cells
and
we
have
the
connections,
so
we
can
at
least
make
comparisons
with
a
full
complement
of
a
connectome,
so
they're
going
to
do
is
they're
going
to
use
a
recent
model
on
theoretical
analysis
that
competes,
the
Dynamics
of
neurobiological
networks
by
focusing
on
the
local
interactions
among
connected
neurons
is
rise
to
the
global
Dynamics
in
an
emergent
way.
So
this
is
where
you
have
these
local
circuits
of
maybe
two
to
three
neurons.
B
B
We
studied
the
Dynamics,
which
resulted
from
stimulating
chemosensory
neuron
Asel
in
a
known
feeding
circuit,
both
an
isolation
and
embedded
in
the
full
connectome.
We
show
that
contralateral
motor,
neurotic
activations
and
ventral
and
dorsal
classes
of
motor
neurons
emerged
from
the
simulations.
So
contralateral
is
the
opposite
side
of
the
organism
or
across
the
midline
of
the
organism.
So
you
have
usually
like
a
right
and
a
left
orientation
of
the
connectome
and
the
contralateral
is
going
from
one
side
of
that
midline
to
another.
B
B
The
connectome
is
generally
up
here
in
the
head,
but
there's
some
neurons
in
the
tail
and
then
there's
a
midline
that
goes
down
the
middle
and
there
are
some
clusters
of
neurons
along
that
they're,
lateral
line
connections
and
things
like
that
that
are
important.
But
this
is
what
the
connectome
looks
like
in
the
adult
and
the
embryo.
B
Of
course,
the
cells
are
clustered
in
one
area
and
they
kind
of
emerge
in
another
area,
they're
different
in
in
the
way
they're
organized
there's
cell
migration,
as
larval
worm
comes
to
be,
and
then
the
larva
worm
has
additional
ecological
adaptations
or
sort
of
facultative
adaptations
as
they
call
them
that
respond
to
different
differences
in
the
environment.
So
the
tower
dower
stage,
for
example,
which
is
between
an
alternate
stage
of
development
between
L2
and
L4.
B
It
has
a
series
of
adaptations
to
very
dry,
very
poor
environmental
conditions,
which
results
in
a
very
thick
cuticle
and
differences
in
the
way
some
of
the
sensory
neurons
are
wired
because
their
inputs
are
different.
So
these
are
this.
Is
you
know?
This
is
the
way
it's
set
up
and
then,
of
course,
we
have
because
we
have
the
head
and
the
tail
and
the
midline.
We
have
this
left
and
right
left
and
right
symmetry,
and
this
is
the
head
of
the
worm
and
we
basically
have
the
Symmetry
in
the
connectome
between
left
and
right.
B
Stands
it's
like
ASE
and
then
L
and
we
also
have
an
aser.
So
we
have
two
cell
types.
One
is
left.
One
is
right:
there's
the
Symmetry
throughout
the
connectome,
sometimes
there's
asymmetry
depending
on
the
neuron,
but
usually
there's
a
left
copy
and
a
right
copy,
and
those
emerge
in
in
the
embryo,
from
cell
division
terminal
cell
divisions
that
result
in
different
neurons
and
then
those
are
put
into
place
in
the
adult
and
they're
wired
up.
B
They
migrate
to
their
perspective
position
in
the
adult
and
then
they're
connected
in
this
way,
where
there's
a
symmetry
for
the
most
part
and
that
there's
this
output
in
the
body
of
a
left
and
a
right
hemisphere,
and
then
they
move
accordingly.
So
you
know
left
neurons
on
the
left.
Side
might
control
right
hand,
movements
neurons
on
the
right
side
might
control
left
hand
movements.
B
So
that's
the
kind
of
thing
we're
talking
about
when
they
talk
about
contralateral
and
asymmetry,
and
things
like
that
Okay.
So
one
interpretation
of
these
results
is
that
there
is
an
inherent
and
we
propose
purposeful,
structural
wiring
to
the
CL
against
connectome
that
has
evolved
to
serve
specific
behavioral
functions.
So
this
is
exactly
what
I
was
talking
about
with
respect
to
some
of
these
circuits
that
there's
mapping
from
a
small
set
of
cells
to
a
larger
set
of
behaviors
or
a
specific
behavior.
B
Now
this
is
not
really
controversial
in
C
elegans,
but
in
larger
organisms
or
organisms
with
maybe
larger
brains,
or
you
know
different
types
of
behaviors,
more
complex
behaviors.
This
is
not.
This
is
not
really
something
that
we
accept
as
easily
and,
of
course,
we
don't
see
that
in
a
lot
of
cases,
C,
elegans
and
and
some
other
organisms
like
anawids,
which
also
have
a
very
small
connectome.
You
see
these
very
simple
behaviors
emerge.
We've
also
done
things
with
Rain
Bird
Vehicles,
which
are
another
example
of
like
a
model.
B
But
this
is
temporal
sequences.
Basically,
these
time-ordered
walks
of
signals
on
graphs.
So
what
they're
doing
is
they're
building
a
graph
like
our
connectome
and.
A
B
A
B
But
there's
also
this
message
passing
aspect
so
whatever,
however,
the
message
is
passed
it
you
know
it's
very
specific,
so
you
have
a
going
to
b
b,
going
to
C
C
going
to
a
that's
a
circuit
where
you
have
a
signal
that
travels
around
this
circuit,
but
with
feedback
when
it
goes
from
C
to
a
and
so
they're
very
specific
orders
that
you
can
have
these
messages
passed
and
a
lot
of
times.
This
is
used
to
reconstruct
graph
neural
network
type
embedding.
So
this
is
some
technique.
B
But,
aside
from
this,
what
we're
doing
here
is
we're
not
just
inferring
a
connection
based
on
some.
You
know,
you
know,
aggregate
message
that
we're
not
defining,
but
we
actually
know
that
this
is
some
sort
of
signal
and
the
signal
then
has
properties
over
time.
B
So
we
found
that
only
five
percent
of
T
seek
are
preserved
between
the
isolated
feeding
Network
relative
to
its
embedded
counterpart.
So
this
is
where
you
have,
in
this
case
they're
talking
about
embedding
in
a
different
way,
they're
talking
about
taking
the
feeding
Network
as
a
single
sort
of
closed
Network
versus
a
network.
That's
connected
to
other
things,
so
any
connectome
is
going
to
be
connected
both
to
itself
to
that
ABC
Clique.
That
I
showed
you
here
and
then
the
other
thing
so
B
is
connected
to
this.
C
is
connected
to
this.
B
A
is
probably
connected
to
other
things:
I'm,
not
showing
those
are
going
to
be
signals
that
come
in
and
out
as
well,
and
so
those
are
going
to
affect
the
sort
of
the
behavior
of
that
little
Clique.
If
you
close
it
off,
if
you
just
consider
a
b
and
c,
it's
a
certain
Behavior,
if
you
add
in
signals
to
different
nodes
from
other
sources,
that's
going
to
affect
it
as
well.
So
when
they
say
embedded,
they
just
mean
this
Clique.
B
The
remaining
95
of
signal,
Wing
Pathways
computed
in
the
isolated
Network,
are
not
present
in
the
embedded
Network.
So
this
means
that
there's
a
huge
difference
between
just
considering
a
clique
and
just
in
considering
the
entire
network.
So
there's
this.
What
they're
saying
is
basically
that
there's
a
submergent
property
that
occurs
from
small
control,
Network
being
embedded
in
a
larger,
connective
and
I.
Think
that's
you
know,
maybe
a
little.
It
may
seem
a
little
hand
wavy
or
obvious,
but
it's
actually
important,
because
we
don't
really
think
about
that.
B
When
we
talk
about
circuits
and
the
literature,
we
usually
talk
about
a
circuit
and
it's
being
responsible
for
X
Behavior
and
the
reason
I
brought
up.
The
whole
thing
about
larger
networks
has
more
complex.
Behaviors
is
because
that's
exactly
what
you
have
in
larger
brains,
you
have
these
nominally,
you
have
circuits,
but
those
circuits
are
connected
to
other
things,
and
so
in
C
elegans,
where
you
have
302,
neurons
and
I.
Think
a
couple
thousand
connections
of
the
most
you
don't
really.
This
problem
isn't
really
very
hard.
B
I
mean
you
know,
you
can
have
business
cliques,
these
circuits
that
we
can
look
at.
We
can
isolate
them.
Anatomically
and
they're
relatively
isolated
in
the
connectivity,
stand
from
a
connectivity
standpoint,
but
then,
when
we
get
to
larger
brains,
we
have
these
things
that
are
more
connected,
more
fully
connected
or
hard
to
discern
as
like
independent
modules
and
it's
hard
to
discern
their
outputs,
because
sometimes
they
have
many
many
outputs,
and
so
this
is
the
problem.
B
B
So
you
know
this
is
something
they
talk
about.
Here
is
the
functional
consequences
of
the
dinome,
so
they
talked
about
the
dyno
and
I
guess
the
dinome
is
like
the
dynamic
aspects
of
the
connectome,
which
is
really
interesting,
because
we
don't
I've
not
heard
that
word
used
before,
but
they
actually
talk
about
it
in
the
paper
here
and
it's
kind
of
their
point.
B
So
this
is
kind
of
going
through
some
of
these
temporal
sequences
going
through
the
cliques,
the
the
sort
of
the
geometric
leaks
and
then
going
through
how
these
things
can
vary
and
then
going
through
their
connections
to
the
larger
Network
and
then
those
temporal
sequences
increase
in
terms
of
their
influence,
because
you
have
more
of
them
coming
in
to
that
little
clique.
B
So
this
is
figure
one.
This
shows
a
reconstruction
of
the
C
elegans
connectome.
So
these
are
the
edges
in
Gray,
they're
they're,
the
specific
neurons
here
colored,
and
so
they
did
some
simulations
here
and
they
show
that
the
number
of
unique
networks-
biking
States,
increases
actually
over
relatively
few
simulation
time.
Steps
So
within
a
thousand
time
steps
they're
able
to
get
a
huge
number
of
unique
Network,
spiking,
States
and
I.
Guess
this
blue
function
here.
If
we
look
at
D,
so
the
full
connectome
is
the
red
curve.
B
The
lattice
is
the
blue
curve.
So
this
is
where
you
get
like
this
difference
between
the
small
circuits
and
the
whole
connector,
but
also
you
know
when
they
arranged
in
the
lattice.
That
means
you're
limiting
the
number
of
connections.
So
a
lot
of
connectomes
are
organized
according
to
sort
of
a
small
world,
or
at
least
a
scale-free
principle,
and
so,
if
you've
ever
heard
of
this
term,
this
is
where
you
have
no
sort
of
characteristic
number
of
nodes
to
something
where
you
have
sort
of
hubs:
hand,
peripheral
nodes.
B
So
it's
like
traveling
in
a
airline
Network
where
you
have
cities
that
are
joined
by
hubs
and
so
getting
from
City
to
city
is
a
very
short
set
of
hops,
as
opposed
to
something
like
a
lattice
where
you
have
a
deterministic
number
of
connections
for
each
node,
let's
say
three,
two
or
three
I
guess
well.
This
has
four.
This
has
three.
It
says
two,
you
know
we
can,
but.
B
B
The
number
of
hops
you
need
to
take
to
go
from
one
side
of
the
network
to
another
is
going
to
be
dependent
on
its
position
in
the
lattice,
rather
than
being
able
to
find
hops
between
hubs
and
then
finding
another
node
on
the
other
side.
So
this
is
why
you
get
this
kind
of
result,
and
if
you
see
the
simulation,
it
extends
out
quite
a
bit
it.
It
finds
a
large
number
of
unique
networks,
biking
States
very
quickly.
B
This
is,
of
course,
where
they're
doing
this,
that
they're
treating
the
connections
as
signals
so
they're,
actually
looking
at
the
number
of
spikes
here
in
the
full
work
in
the
full
Network
and
then
the
randomized
full
Network
and
then
they're
also
looking
at
these
simulation
time,
steps
for
these
T
seeks.
So
these
are
the
T
seeks.
So
this
is
a
temporal
sequence
from
Asel
to
the
VB
and
DB
classes
of
neuron.
These
are
where
you
have
neuron
number
and
simulation
time
steps.
B
So
I
think
that's
enough
for
that
paper.
Now,
I
want
to
talk
about
this
paper,
which
is
moving
on
to
Alternative
neural
systems,
so
this
is
actually
getting
to
something
we
talked
about
with
respect
to
sulfates
before,
and
that
is
what
is
a
neuron.
And
so
when
we
talk
about
C
elegans
connector
and
we
talk
about
a
simple
system
relatively
simple
system
where
we
have
neurons
and
we
have
X,
you
know
axons
and
synapses,
like
we
do
in
say
the
human
brain.
But
we
have
you
know
these
sort
of
stereotypical
behaviors.
B
We
have
a
much
stronger
link
between
neural
circuits
and
behaviors
and
so
forth.
But
then
the
question
is
what
is
a
neuron
and
what
is
a
nervous
system?
B
And
so
this
actually
goes
further
to
say
that,
like
you
know
you
don't
even
maybe
don't
even
need
a
a
connectome
to
to
produce
some
of
these
behaviors.
So
this
is
an
interesting
paper
that
focuses
on
tinafore's,
sponges
and
plaque.
Placazones
and
so
in
these
organisms
they
have
something
called
a
nerve
net.
They
also
have
you
know
they
express
the
proteins
for
synapses,
but
they
don't
actually
have
the
kind
of
synapses
that
you
would
find
in
C,
elegans
or
in
humans.
So
this
is
one
different.
B
B
The
brief
answer
is
no
for
early
Divergent
lineages
from
a
common
ancestor
from
a
nervous
common
ancestor
of
all
animals,
so
C
elegans
and
humans-
and
you
know
a
lot
of
animals
are
in
this
part
of
the
tree
of
life,
where
you
have
nervous
systems
where
you
have
these
specific
nervous
systems
that
we're
familiar
with,
and
then
there
are
other
parts
of
the
tree
of
life
and
animals.
Where
you
don't
have
this,
you
have
different
approaches
to
building
a
nervous
system,
so
there
are
some
really
interesting
types
of
they
call
them
neuroid
type
integrative
systems.
B
So
there
are
a
lot
of
different
types
of
these
in
animals
and,
and
one
way
is
of
course
the
way
we're
familiar
with
this
other
way
is
where
you
have
say,
cone
jellies
with
unique
synapses
there's.
So
one
of
these
is
a
subset
of
neural
Nets
and
comb
jellies
with
unique
synapses.
The
second
lineage
is
the
well-known
Nigeria
plus
bioateria.
Of
course,
this
is
where
we
are.
B
The
two
others
are
non-synaptic
neuroid
systems
and
sponges
and
placozones.
So
this
is
non-synaptic,
so
you
have
some
with
unique
synapses,
some
with
just
regular
synapses
that
we're
familiar
with
and
then
some
with
non-synaptic
systems
that
are
also
neural
in
nature.
So
by
integrating
scr,
rna-seq
data
and
microscopy
data,
we
revise
the
definition
of
neurons,
a
synaptically
coupled
polarized
and
highly
heterogeneous
secretory
cells
at
the
top
of
Behavioral
hierarchies
with
learning
capabilities.
So
that's
a
mouthful
that
is
so,
let's
see
if
we
have
so.
B
The
first
thing
is
the
first
part
of
this
definition
is
synaptically
coupled
polarized
and
highly
heterogeneous
secretory
cells.
So
we
talked
about
synaptically
coupled
so
you
know,
they're
coupled
in
some
way
actually
I
think
synaptic
we
coupled
might
be
a
little
bit
because
they
did
talk
about
how
there's
either
these
non-synaptic
neuroid
systems,
but
you
do
have
to
have
connections
between
them
all
right,
so
the
network
has
to
have
connections.
B
B
B
Another
way,
of
course,
is
through
the
typical
synaptic
connection
or
some
sort
of
connection
between
them,
a
physical
connection,
which
is
usually
a
chemical
connection,
so
you
have
a
synapse
here.
You
know
you
have
maybe
it's
an
axon
and
synapse
pairing.
It
could
be
something
else.
The
Third
Way,
of
course,
then,
is
secretions,
and
so
this
is
where
the
cell
secretes
proteins
out
of
its
cell
body
and
into
the
local
environment.
B
So
these
cells
here,
depending
on
their
distance
from
it,
we'll
get
a
dosage
of
whatever
is
being
secreted
from
the
cell
and
hopefully
it'll
respond
in
kind.
So
there
are
a
lot
of
peptide
systems
or
peptid
urgent
signaling.
That
works
in
a
way
like
this,
where
it's
gets.
It
gets
secreted
or
released
into
the
environment
and
it
gets
uptaken
by
another
cell,
sometimes.
D
B
Restricted
to
something
you
find
in
the
synapse
in
like
in
C
elegans,
we
have
something
called
juxtaprint
signaling,
which
is
where
signals
get
sent
out
with
in
protein
form
and
they
get
taken
up
by
its
neighboring
cells
and
there's
the
cell
cell
signaling.
That
happens,
there's
also
a
paracrine
signaling,
which
happens
at
longer
distances,
but
the
secret
home
is
interesting
because
there's
a
lot
of
stuff
that
gets
pumped
out
of
the
cell
and
into
the
local
micro
environment
of
the
of
the
cells
or
the
nervous
system.
B
That's
very
interesting,
and
this
is
of
course
an
alternate
way
of
building
a
nervous
system,
it's
not
as
efficient
as
the
synapse
and
it's
not
even
as
efficient
as
the
Gap
Junction.
But
there
is
this
aspect
of
where
you
have
the
signaling
and
that's
the
important
thing
is
the
signaling,
so
I
want
to
label
these.
This
is
a
gap.
Direction
This
is
electrical,
B
is
a
synapse,
it's
chemical
and
then
C
is
the
secretum,
which
is
also
chemical.
B
So
this
physiological
Def
and
then
of
course,
we
have
behavioral
hierarchies,
which
means
that
you
know
you
have
behavioral
control.
You
have
control
of
some
complex
Behavior,
you
have
a
set
of
neurons,
for
you
know
one
function
versus
another
function
and
you
can
compose
them
with
learning
capabilities,
which
means
that
these
can
establish
learning.
They
can
respond
in
kind.
You
can
have
like
heavy
in
rules
even
in
a
secret
home.
You
can
have
other
types
of
connections
where
the
connections
are
strengthened
over
time
and
that's
the
basis
for
learning.
B
At
least
we
think
it
is
this
physiological,
not
phylogenetic
definition.
So
they're,
not
thinking
about
this
in
terms
of
the
Tree
of
Life
they're,
just
saying
that
this
is
the
physiology,
the
things
that
these
physiological
systems
have
in
common
separates,
true
neurons
from
non-synaptically
and
GAP
Junction,
coupled
integrative
systems.
So
this
is
where
you
have.
This
definition
separates
us
out
from
sort
of
true
neurons
and
non-synaptic
GAP.
We
talked
about
this
executing
more
stereotypic
Behavior,
so
this
this
definition
actually
can
differentiate
these.
B
So
this
is
where
they're
just
kind
of
like
saying
no
true
neuron.
Does
this
or
a
true
neuron
does
this
is
sort
of
like
a
notary,
Scotsman
argument,
but
okay
growing
evidence
supports
the
hypothesis
of
multiple
origins
of
neurons
and
synapses.
So
these
happen
multiple
places
in
the
tree
of
life
or
they
say,
originate
their
as
many
non-violitarian
and
bioletary
neuronal
classes,
circuits
or
systems
are
considered
functional
rather
than
genetic
categories.
So,
like
I
talked
about
earlier,
you
know
you
have
these
genes
that
get
expressed.
You
can
get
a
gene
expression
profile.
B
Those
don't
tell
you
necessarily
what
the
function
is.
It
just
tells
you
what
genes
are
overregulated.
The
functional
definition
is
different.
It
could
be
related
to
the
things
that
it's
expressing,
but
those
aren't
sufficient
to
explain
the
function,
those
those
classes
of
things
that
are
upregulated
or
down
regulated
for
that
matter.
It's
only
a
clue.
It's
only
a
marker
of
maybe
like
Identity
or
function,
but
it
doesn't
tell
you
anything
about,
like
the
causes
of
it
or
the
white
defines
what's
going
on
there.
B
It's
kind
of
a
loose
Association,
sometimes
a
correlation,
but
you
have
to
keep
that
in
mind,
because
sometimes
you
know
when
you
identify
something
from
transfer.
Atomics,
it
can
kind
of
conflict
with
some
of
the
things
we
find
from
functional
studies
or
from
well.
They
include
microscopy
here,
but
there's
a
lot
that
is
conflicting
between
transcriptomic
profiles
and
the
sort
of
things
we
can
see
under
a
microscope
in
terms
of
function
or
in
terms
of
structure.
B
So
thus
many
non-bioletary
and
bioletarian
Rural
classes,
circuits
or
systems
are
considered
functional
rather
than
genetic
categories
composed
of
non-homologous
cell
types.
In
summary,
little
explored
examples
of
convergent,
neuronal
evolution,
representatives
of
early
branching
metazones
provide
conceptually
novel,
microanatomical
and
physiological
architecture
of
Behavioral
controls
in
animals
with
prospects
of
neural
engineering
and
synthetic
biology.
So
this
is
where
we
can
learn
from
these
systems.
We
can
maybe
mimic
them,
we
can
build.
B
You
know
something
that
we
can
modify
them
and
see
how
they
work,
where
we
can
build
our
own
system
without
all
of
its
features.
We're
probably
not
all
of
its
features.
But
you
know
this
is
a
nice
review
kind
of
goes
through
this
argument.
They're
making
an
argument
here,
they're
kind
of
making
the
diff
you
know
exploring
these
different
classes
of
neuronal
type
of
different
types
of
normal
architecture.
They
say:
George,
Streeter
who's
wrote
a
book
on
neuro
Evolution.
So
there's
this.
C
B
Actually
an
example
of
plural
brachia,
where
you
have
the
subepithelial
neural
net,
so
you
have
these
cells
in
the
self-epithelium,
it's
a
network,
and
so
it's
not
really
like
a
centralized
nervous
system,
it's
sort
of
distributed
throughout
the
body
of
the
organism
and
it's
it's
integrated
into
the
epithelium.
So
it's
it's
sort
of
underneath
epithelial
layer,
but
it's
it's
all
throughout
the
apathy
oil
there,
and
so
this
helps
with
the
function
of
this
organism,
because
this
organism
needs
a
sort
of
distributed
processing.
B
Neuronal-Like
cells
throughout
in
in
and
around
the
sub
epithelial
layer,
and
then
you
have
you
know
the
furrow
here
at
the
bottom,
and
so
this
is,
and
these
are
receptors
kind
of
coming
out
of
the
body.
So
you
have
these
tentacles
coming
out
and
they're
bringing
their
signals
into
right
directly
into
these
nuts,
whereas
in
in
in
other
organisms,
you
have
a
system
of
bringing
them
in.
You
have
relays,
and
things
like
that.
You
have
a
centralized
nervous
system.
B
You
need
to
bring
things
in,
and
so
that's
that's
the
difference,
and
there
may
be
the
reasons
why
they
have
this
sort
of
distributed
nervous
system.
I
mean
we
have
to
some
extent
a
distributed,
peripheral
nervous
system,
but
that's
a
different
system,
because
it's
connected
to
a
centralized
system,
it's
largely
for
sort
of
external
peripheral
processing
and
moving
your
arms
and
your
legs,
and
things
like
that
and
having
that
kind
of
resolution
of
the
body
as
a
reference
point.
But
this
is
actually
where
it's
distributed
throughout
the
body.
B
B
The
abstract
reads:
during
brain
development:
billions
of
axons
must
navigate
over
multiple
spatial
scales
to
reach
specific
neural
targets,
so
this
is
where,
in
brain,
development
axons
have
to
find
other
neurons.
So
there's
this
idea
of
axon
finding
and
there's
this
axon
cone,
which
allows
it
to
detect
signals
and
find
other
neurons
to
connect
to,
and
so
this
is
definitely
where
you
have
a
know.
B
B
So
it
looks
kind
of
like
this,
where
you
have
like
cells
that
are
maybe
just
becoming
neurons,
and
then
they
have
these
processes
that
kind
of
come
out
and
again
you
know
they
have
these
chemical
signals
that
are
secreted
out
from
the
cell,
so
it's
sending
out
signals
that
decay
at
a
certain
rate
as
they
get
away
from
it,
and
these
axons.
These
processes
are
sticking
out
and
they're
kind
of
moving
along
these
gradients
of
chemicals.
B
B
The
key
here
is
that
the
individual
cells
are
secreting,
proteins
or
secreting
signals
for
other,
like
for
axons
to
pick
up,
so
they
have
this
axon
cone,
which
is
sort
of
like
the
end
of
the
axon
and
controls
the
growth
rate
and
where
it's
going
in
its
direction.
So
this
is
how
axons
find
Targets,
and
this
is
something
that
in
evolution,
changes
because
you
can
have
Target
tissues
in
different
places
in
the
organism.
B
Let's
call
this
a
Target
tissue
and
you
can
have
motor
neuron
axons,
for
example,
that
go
down
and
they
have
to
innervate
the
Target
tissue.
They
have
to
find
the
target
tissue
and
if
the
target
tissue
isn't
there
say,
for
example,
there's
a
like
a
tail
and
the
motor
motor
neuron
sends
out
an
axon,
it
innervates
the
tail,
and
then
that
controls
the
tail.
B
What
will
happen
is
that
the
axon
won't
migrate
out,
but
it
won't
find
a
Target,
so
it
either
finds
another
Target
tissue
innervate,
which
means
that
you
have
plasticity
here
in
evolution.
So
it's
sort
of
an
evodivo
thing
where
it
finds
a
new
Target
tissue
or
the
cell
dies,
and
simply
you
lose
motor
neurons
for
that
Target
tissue
and
can't
reuse
it.
So
there
are
a
lot
of
scenarios,
and
this
is
all
spelled
out
in
the
Streeter
book
that
I
mentioned
earlier.
B
This
is
you
know
where
a
briefly,
you
know,
just
to
recap:
axons
are
sent
out.
They
find
their
targets
either
it's
a
cell
or
a
tissue.
B
If
the
target's
out
there,
the
axon
Withers
and
then
the
cell
may
die
if
it's
not
connected
to
anything-
and
this
is
a
lot
of
the
processes
of
development
that
are
involved
in
evolutionary
changes
in
phenotype
in
like
morphology
or
sometimes
just
you
know,
because
it
doesn't
have
a
tail,
and
so
you
know
it
just
it's
a
developmental,
this
developmental
defect
and
doesn't
have
a
tail.
So
the
motor
neurons
aren't
there
too,
and
you
can
actually
look
in
a
Define
mutant
to
see
this
sort
of
thing.
B
So
this
is
this
is
all
what
they're
talking
about
here
with
axons
and
navigating
to
their
targets
and
so
build
the
processing
circuits
that
generate
the
intelligent
behavior
of
animals.
However,
the
limited
information
capacity
of
the
zygotic
genome
puts
a
strong
constraint
on
how
and
which
axonal
roots
are,
can
be
encoded.
So
this
you
know
the
genome
is
actually
playing
the
role
of
building
these
proteins
from
transcriptomes
that
actually
allow
for
these
signals
to
be
sent
out
and
then
for
this
communication
to
happen.
B
So
it
all
goes
back
to
what
the
genome
is
able
to
encode.
If
I
can
encode
the
signals,
if
I
can
encode
it
at
a
you
know
in
in
time,
or
you
know,
at
a
enough
of
it
is
produced,
then
you
can't
have
the
system
doesn't
work,
so
these
axonal
routes
are
con,
controlled,
ultimately
by
The
zygotic
genome.
We
propose
and
validated
mechanism
of
development
that
can
provide
an
efficient
encoding
of
This
Global
wiring
task.
B
So
a
lot
of
people
have
approached
this
wearing
task.
As
a
matter
of
optimizing,
the
wiring,
so
the
one
way
you
can
do
this
to
understand
the
wearing
patterns
is
to
optimize
it
with
respect
to
the
topology.
So
you
find
the
shortest
paths
where
you
find.
You
know
the
least
overlapping
paths,
and
you
assume
that
that's
the
way
that
this
system
will
evolve
because
it's
efficient
and
so
in
some
ways,
that's
true,
but
in
other
ways
you
know
we
have
these
other
constraints
that
play
a
role
in
how
these
connectomes
are
constructed.
B
So
these
are
daughters,
like
basically
daughter
cells,
that
divide
at
the
same
time.
So
in
other
words,
if
we
think
back
to
our
lineage
tree,
we
know
that,
for
example,
if
you
have
a
mother
cell,
which
is
up
here
at
the
top,
should
I
have
that
already
all
right.
That
Mother
cell
then
produces
two
daughter
cells
or
more
sometimes,
there's
proliferation.
Here
those
daughter,
those
mother
cells
will
pass
on
gene
expression
to
its
daughter
cell.
Sometimes
it
passes
on
the
RNA
in
the
compartment.
B
The
daughter
cells
here,
which
are
maybe
more
mature
cells-
the
constraint,
is
set
here
at
the
top
and
it
constrains
what
the
daughters
are
doing,
but
it
also
sometimes
mRNA
gets
carried
over
compartmentally
in
the
cells
so
that
you
know
this
is
important
like
in
labeling
the
cells.
B
But
it's
also
an
important
point
in,
like
you
know
what
gets
upregulated,
what
gets
done
regulated
things
like
that,
so
there
definitely
constraints
based
on
Cell
lineage,
and
so
that's
what
they're
talking
about
here
and
this
induces
a
global
hierarchical
map,
nested
regions,
each
marked
by
the
expression
profile.
That's
common
in
progenitor
population.
B
B
So
there's
the
staged
route
of
cells
and
there
are
descendants-
and
this
can
be
this
usually
happens
in
space.
There
are
different
ways
that
this
happens
like
they're
different
patterns
of
birth,
birth
of
cells
and
they
migrate
in
a
certain
pattern
and
the
growth
cones
of
the
axons
can
follow
those
growth
patterns
because
they're,
actually,
you
know,
set
up
chemically
in
that
way.
They're
consistent
chemically
and
then
you
know,
there's
a
there's
a
pattern
there
for
the
growth
cone
to
follow.
B
We've
analyzed
gene
expression,
data
of
developing
in
adult
Mouse
brains,
published
by
the
Allen
Institute
of
brain
science.
So
the
brain
Allen
Institute
has
a
really
nice
set
of
mouse
brain
atlases,
which
show
like
not
only
the
anatomy
and
high
detail,
but
the
gene,
expression,
maps
and
spatial
in
their
spatial
distribution
in
high
detail.
So
they've
done
this
for
I,
think
the
entire
Mouse
brain
or
a
number
of
different
Mouse
brains.
So
they've
done
this
for
development
and
for
the
adults.
B
This
is
a
good
resource
if
you're
interested
in
this
more
and
found
them
consistent
with
our
simulations
gene
expression
induced
partitions
and
brain
the
brain,
the
global
spatial
hierarchy
of
nested,
contiguous
regions
that
is
stable
from
at
least
embryonic
day,
11.5
to
postnatal
day
56..
So,
basically,
what
happens
is
that
you
have
the
brain
as
it's
developing
in
at
least
in
Mouse.
You
have
these
sort
of
hierarchical
sections
where
you
know.
Basically,
development
is
stable.
It
allows
for
development
to
proceed
without
just
sort
of
this
problem
of
search.
B
So,
like
you
know,
the
axon
doesn't
need
to
search
every
possible
combination.
It
just
needs
to
find
its
way
through
the
predefined
path
and
that's
makes
it
a
lot
easier
for
these
processes
to
happen
properly,
and
so
you
know
we
talk
about
how
development
is
a
process,
that's
very
regular,
most
of
the
time,
and
it's
amazing
how
that
happens,
and
some
of
how
that
happens
is
to
have
these
constraints
on
what
the
developmental
system
actually
needs
to
do
to
achieve
a
stable
phenotype.
B
We
use
this
experimental
data
to
demonstrate
that
our
axonal
guidance
algorithm
is
able
to
robustly
extend
Arbors
over
long
distances
specific
targets.
So
your
axon
is
like
a
tree.
You
know
it
arborizes,
and
you
know
that
sort
of
thing.
So
it's
doing
this
as
well
as
like
finding
targets
and
that
these
connections
result
in
a
qualitatively
plausible
connectome,
so
they're
actually
modeling
this
and
showing
that
it
does
produce
the
kind
of
a
result
that
we
would
see
in
a
real
life,
biological
connective,
so
they're
doing
simulations
here.
B
We
conclude
that,
paradoxically,
cell
division
may
be
the
key
to
uniting
neurons
in
the
brain.
So
this
is
the
Paradox
of
cell
cells
divide.
They
actually
keep
this
the
brain
together.
So
this
is
an
interesting
paper.
I
hadn't
thought
about
it
in
this
way
before,
but
this
is
something
they
call
constructive
connectomics.
B
There
are
two
aspects
of
the
construction
problem:
the
generation
from
a
few
precursors,
the
vast
number
of
various
types
of
neuron
that
comprise
the
brain
and
the
process
whereby
these
neurons
connect
to
one
another.
So
it's
generating
neurons
and
connecting
them.
The
process
of
generation
is
well
understood.
We
have
lineage
Maps
fate,
maps
and
lineage
trees
that
we
can
use.
But
you
know
this
process
of
connection
is:
is
the
harder
problem,
so
this
is
this
is
what
they're
talking
about
and
so
they're
using
these
they
have
these.
B
They
have
their
lineage
tree,
hair
cells,
proliferation,
axon
navigation,
and
then
you
have
this
connectome
as
once
everything's
connected
up.
So
that's
the
problem
and
that's
their
solution.
Is
this
constructive,
connectomics,
so
I
think
those
were
a
nice
set
of
papers?
Thank
you
for
watching
and
I
hope.
You
learned
something.